Effect of Screening and Adjuvant Therapy on Mortality from Breast Cancer

The new england journal of medicine original article Effect of Screening and Adjuvant Therapy on Mortality from Breast Cancer Donald A. Berry, Ph.D., Kathleen A. Cronin, Ph.D., Sylvia K. Plevritis, Ph.D., Dennis G. Fryback, Ph.D., Lauren Clarke, M.S., Marvin Zelen, Ph.D., Jeanne S. Mandelblatt, Ph.D., Andrei Y. Yakovlev, Ph.D., J. Dik F. Habbema, Ph.D., and Eric J. Feuer, Ph.D., for the Cancer Intervention and Surveillance ing Network (CISNET) Collaborators* abstract From M.D. Anderson Cancer Center, Houston (D.A.B.); the National Cancer Institute, Bethesda, Md. (K.A.C., E.J.F.); Stanford University, Stanford, Calif. (S.K.P.); the University of Wisconsin Madison, Madison (D.G.F.); Cornerstone Systems, Lynden, Wash. (L.C.); Dana Farber Cancer Institute, Boston (M.Z.); Georgetown University, Washington, D.C. (J.S.M.); the University of Rochester, Rochester, N.Y. (A.Y.Y.); and Erasmus University Medical Center, Rotterdam, the Netherlands (J.D.F.H.). Address reprint requests to Dr. Berry at the Department of Biostatistics and Applied Mathematics, M.D. Anderson Cancer Center, Unit 447, 1515 Holcombe Blvd., Houston, TX 773, or at dberry@mdanderson.org. *The CISNET collaborators are listed in the Appendix. N Engl J Med 25;353:1784-92. Copyright 25 Massachusetts Medical Society. background We used modeling techniques to assess the relative and absolute contributions of screening mammography and adjuvant treatment to the reduction in breast-cancer mortality in the United States from 1975 to 2. methods A consortium of investigators developed seven independent statistical models of breast-cancer incidence and mortality. All seven groups used the same sources to obtain data on the use of screening mammography, adjuvant treatment, and benefits of treatment with respect to the rate of death from breast cancer. results The proportion of the total reduction in the rate of death from breast cancer attributed to screening varied in the seven models from 28 to 65 percent (median, 46 percent), with adjuvant treatment contributing the rest. The variability across models in the absolute contribution of screening was larger than it was for treatment, reflecting the greater uncertainty associated with estimating the benefit of screening. conclusions Seven statistical models showed that both screening mammography and treatment have helped reduce the rate of death from breast cancer in the United States. 1784

the cancer intervention and surveillance ing Network (CISNET) is a consortium of investigators sponsored by the National Cancer Institute whose purpose is to measure the effect of cancer-control interventions on the incidence of and risk of death from cancer in the general population. This report of the CISNET Breast Cancer Working Group provides estimates of the contributions of screening mammography and adjuvant treatment to the reduction in the rate of death from breast cancer among U.S. women from 1975 to 2. In 1975, the rate of death from breast cancer among women 3 to 79 years of age, adjusted for age to the 2 population, was 48.3 deaths per 1, women. By 199, the rate had increased slightly to 49.7 per 1,, but then fell to 38. per 1, by 2, a decrease of 24 percent from 199. 1 Similar reductions have been observed in other countries. 2 Likely explanations for these reductions are early detection by mammographic screening and advances in treatment. Screening can reduce the rate of death from breast cancer only when followed by treatment. Tumors that are detected by screening before they metastasize can be cured by surgery, and breast cancer detected at an early stage of metastasis can be effectively treated by chemotherapy. 3 Currently, most organizations recommend that women begin to undergo screening mammography in their 4s. 4-6 By the year 2, about 7 percent of women over the age of 4 years reported undergoing mammography in the previous two years. 7 Controversies over the value of screening mammography arose from questions regarding the quality of the randomized trials that evaluated the effectiveness of this approach. 6,8-1 Moreover, the most recent estimates of benefits of screening from some of these trials are substantially lower than earlier estimates. 9 Even if screening trials provided undisputed evidence of a reduced rate of death from breast cancer, it is unclear how such a finding would translate to the general population. The use of adjuvant chemotherapy and tamoxifen for all stages of breast cancer has also increased. Randomized clinical trials have demonstrated survival benefits associated with the use of adjuvant therapies, with estimated reductions in the annual odds of death ranging from 8 percent to 28 percent, depending on the type and duration of therapy, the age of the patients, and the characteristics of the tumor. 11-13 However, the extent to which these benefits translate to the population outside the controlled conditions of clinical trials is unknown. Furthermore, whether advances in adjuvant treatment have diminished or enhanced the value of screening has not, to our knowledge, been analyzed previously. Knowing whether screening, adjuvant therapy, or both have contributed to the recent reductions in the rate of death from breast cancer is important. Since no single national registry contains all the data needed to evaluate the effect of these interventions at the population level, a way of integrating relevant information from separate population databases is required. The National Institutes of Health selected seven groups to develop models of mortality from breast cancer in the United States. These models made use of the best information available from the period 1975 to 2 as common sources of data but applied different modeling approaches. The results of these various approaches differed, but there were points of agreement in the overall conclusions. methods The National Institutes of Health used a competitive peer-review process to choose seven groups to model the effect of screening and treatment on trends in the incidence of and rate of death from breast cancer: the Dana Farber Cancer Institute, Boston (model D); Erasmus University Medical Center, Rotterdam, the Netherlands (model E); Georgetown University, Washington, D.C. (model G); the M.D. Anderson Cancer Center, Houston (model M); Stanford University, Stanford, California (model S); the University of Rochester, Rochester, New York (model R); and the University of Wisconsin Madison, Madison (model W). 14 The joint analysis was designed by the CISNET consortium of 43 investigators (listed in the Appendix). The seven groups worked independently to develop their models but interacted as a consortium to investigate shared problems and to facilitate comparisons by developing uniform reporting structures. These comparisons allowed the modelers to identify differences in approaches and underlying assumptions and helped identify errors, but the general modeling approach and model structures were not modified to achieve consistency among the seven groups. The National Cancer Institute prepared databases of the information used in all seven models; individual groups collected data required for their particular models. Data used by all seven groups are 1785

The new england journal of medicine publicly accessible, and additional detail related to the variables can be found on the CISNET Web site (http://cisnet.cancer.gov/). The consortium s analyses relied on the incidence of breast cancer as reported by the Surveillance, Epidemiology, and End Results (SEER) program and the rate of death from breast cancer as reported by the National Center for Health Statistics (NCHS). Because neither SEER nor NCHS contains information on screening history or mode of detection and SEER underreports the use of adjuvant therapy, 15 we incorporated additional databases concerning uses of screening and treatment and their efficacy in the population. In the first phase of our collaboration, we jointly developed and agreed on a set of common, or base-case, variables to make possible comparisons of the models. The base-case variables and the data sources that were used to estimate them are listed in Table 1. Base-case variables include the background incidence of breast cancer in the absence of screening; the use of adjuvant treatment according to the patients age, receptor status, calendar year, and tumor stage; and the risk of death from other causes. The background trend was estimated from the Connecticut Tumor Registry and SEER with the use of an age-period cohort model. 16 Cohort-specific screening patterns were estimated from data on the percentage of the population that had ever undergone mammography, as reported in the National Health Interview Survey 7,17 and the screening patterns from population-based data on the use of mammography collected by the Breast Cancer Surveillance Consortium for the period 1994 to 2. 18 We used breast-cancer specific survival curves derived from SEER for patients who received a diagnosis of breast cancer from 1975 to 1979 to estimate the rate of death among patients who did not receive chemotherapy or tamoxifen. The birthcohort specific rate of death from any cause, obtained from the Human Mortality Database, 19 was adjusted to exclude deaths from breast cancer. Table 2 summarizes the base-case variables and other distinguishing characteristics of the models. Six of the seven groups estimated the benefit derived from screening and treatment with the use of a model of the natural history of invasive breast cancer, with three groups (models E, G, and W) also modeling progression from ductal carcinoma in situ to invasive disease. These three groups included ductal carcinoma in situ as an early, precursor stage of invasive breast cancer. The other models considered only invasive cancer. M did not use a model of the natural history of the disease but used Bayesian updating of model variables including the results of treatment trials to estimate the contributions of the various interventions. Group M simulated screening and breast-cancer histories of women who were alive in 1975. The model allowed for the possibility that patients with tumors detected by screening would survive longer than patients whose cancer was detected on the basis of symptoms, even in the case of tumors at the same stage in both settings. In all models except model R, patients with breast cancer either received adjuvant treatment (tamoxifen, chemotherapy, or both) or did not receive adjuvant treatment, depending on the year, the age of the patients, tumor stage, and estrogenreceptor status. A survival benefit of tamoxifen was applied only to patients with estrogen-receptor Table 1. Base-Case Variables. Variable Data Set* NHIS SEERPOC BCSC SEER 9 Connecticut Tumor Registry Background incidence of breast cancer + + Dissemination of mammography + + Dissemination of treatment + + Human Mortality Database Death from other causes + + Breast-cancer survival in 1975 + Breast-cancer prevalence in 1975 + + NCHS * NHIS denotes National Health Interview Survey, SEERPOC SEER Patterns of Care, BCSC Breast Cancer Surveillance Consortium, and SEER 9 SEER Nine Registries. A plus sign indicates that the data set was used to estimate the variable in question. 1786

Table 2. Specifications of the s.* Characteristic D E G M S R W Variables Total no. 9 21 12 6 1 11 4 No. calibrated according to observed mortality 4 2 No. calibrated according to observed incidence of breast cancer 3 5 2 2 3 38 Type of model Simulation + + + + Analytic + Both + + Aspects included in natural-history modeling Explicit natural-history model + + + + + + Progression From preclinical to clinical disease (unstaged) + + + + From ductal carcinoma in situ to local to regional to distant disease (staged) + + + + Tumor growth (exponential or Gompertzian) + + + + Direction of model Forward, from onset + + + + Backward, from clinical incidence + + Diameter of tumor when fatal + Stage at diagnosis AJCC stage + +** + SEER historical stage + + + + Covariates for sensitivity of screening mammography Age at screening + + + + + Calendar year (or birth cohort) + + + + Tumor size + + + + Breast density + Preclinical stage at screening + + Covariates for breast-cancer survival after diagnosis Age- and stage-specific survival curves + + + + Age-, stage-, and size-specific survival curves + + Cure (or no cure) based on tumor-diameter threshold + + Lead-time guarantee (i.e., death from cancer not allowed during lead time) + + + + + Allowed survival benefit in addition to stage shift + + + + Covariates for treatment benefit after diagnosis (independent of screening) Estrogen-receptor status + + + + + Age + + + + + + Year + * D denotes Dana Farber Cancer Institute, E Erasmus University Medical Center, G Georgetown University, M M.D. Anderson Cancer Center, S Stanford University, R University of Rochester, W University of Wisconsin Madison, and AJCC American Joint Committee on Cancer. A plus sign indicates that the model had the characteristic in question. Posterior distributions were found with the use of Bayesian rejection sampling. The natural-history variables were estimated with the use of incidence data from the Two-County Swedish Study, 2 not SEER data. This number was estimated from the Canadian National Breast Screening Studies CNBSS I and II 21,22 and calibrated according to SEER incidence data. This model did not include in situ breast carcinoma. Preclinical onset was determined through the selection of sojourn time from an exponential distribution. **The stage was driven by size alone, estimated with the use of AJCC data through the calibration of nodal status and metastasis status to basecase stage-distribution variable. 1975 Survival estimates were modified by a drift term. If not cured, disease was assumed to progress to a late stage, with survival characteristics of SEER distant disease. 1787

The new england journal of medicine positive tumors and depended on the duration of treatment (two or five years). The estimated relative-risk ratios for death from any cause were.82 for two years of tamoxifen therapy and.72 for five years of therapy. 13 The relative-risk ratio for chemotherapy depended on the age at detection:.73 for women younger than 5 years of age,.86 for women 5 to 59 years of age, and.92 for women 6 years of age or older. 11 The University of Rochester group (Group R) used SEER data to calibrate their model according to the effect of treatment and the change in survival over time, controlling for age, tumor size, and clinical stage. Thus, the treatment variable in model R included not only adjuvant therapy but also the possibility that surgical and radiation-therapy procedures and patient care more generally resulted in prolonged survival for patients with breast cancer. All the models measured the contribution or effect of various factors, such as the use of screening and the reduction in the relative-risk ratio for death owing to five years of tamoxifen therapy. The modelers had to estimate the effect of these factors. Three models (models M, R, and W) were calibrated according to the observed rate of death from breast cancer in the United States, in that they gave more weight to data that produced results consistent with the observed rates from 1975 to 2. The other four groups did not fit variables on the basis of the observed rate of death from breast cancer in the United States but instead relied on a variety of data sources and processes to derive mortality trends. More detailed descriptions of the individual models are available in the Supplementary Appendix (available with the full text of this article at www.nejm.org) and at http://cisnet.cancer.gov/ resources/. 23 To produce estimates of the effects of screening and adjuvant therapy from 1975 to 2, the models were used to simulate trends in the incidence of and rate of death from breast cancer under four scenarios: no screening and no adjuvant therapy; use of base-case screening, but no adjuvant therapy; no screening, but base-case use of adjuvant therapy; and base-case screening and base-case use of adjuvant therapy ( base-case corresponds to the estimated actual use of screening and adjuvant therapy from 1975 to 2). The models estimated the contribution of screening in the absence of treatment by comparing the first and second scenarios, the effect of treatment in the absence of screening by comparing the first and third scenarios, and the effect of both screening and treatment by comparing the estimated individual contributions with the combined contribution as estimated by comparing the basecase (fourth) scenario with the first scenario. The models provide estimates of absolute reductions in the rate of death in 2 owing to the use of tamoxifen, chemotherapy, these two treatments combined, screening, and all these factors combined. We present and plot the seven pairs of estimates (declines in the rate of death from breast cancer due to screening and adjuvant therapy) that were produced by the models. For the purposes of representing the uncertainty in these estimates, we regarded these seven pairs to be a sample from a larger population of possible model results. Using a statistical method called kernel estimation, 24 we estimated the distribution of this larger population. We used a bivariate normal kernel with a standard deviation and correlation taken from the data plot of the seven estimates. results Figure 1A shows the use of screening (annual, every two years, irregular, and never) among women who were at least 4 years of age in calendar years 1985, 199, 1995, and 2. 25 The use of screening increased dramatically during this period. The use of adjuvant treatment depended on the calendar year, the age of the patients, the tumor stage, and estrogen-receptor status. 26 These data were derived from SEER treatment information adjusted for underreporting as assessed from SEER patternsof-care studies. 15 Figure 1B shows representative curves describing the use of treatment for nodepositive stage II or stage IIIA disease among women 5 to 69 years of age. The percentage of women who received chemotherapy increased from essentially percent in 1975 to about 8 percent in 2. Similarly, the use of tamoxifen increased from essentially percent in 1975 to about 5 percent in 2; this drug was used mainly in patients with estrogen-receptor positive tumors. Figure 2A shows the rates of death from breast cancer from 1975 to 2 as estimated by the seven models, in comparison with the actual rate in the United States. s that were calibrated according to the actual mortality rate (models M, R, and W) tend to fit better than those that were not calibrated in this manner. However, the shapes of the seven curves are similar. Moreover, Figure 2A shows that all models predict similar proportional reductions in mortality from the combination of screening and adjuvant therapy. Figure 2B shows the age-adjusted 1788

A Percentage Undergoing Screening Mammography 1 75 5 25 Never Irregular Every 2 yr Annual 1985 199 1995 2 Year B Percentage Receiving Treatment 6 5 4 3 2 Multiagent chemotherapy only Both 1 Tamoxifen only 197 198 199 2 Year Figure 1. Changes in the Pattern of Use of Screening Mammography among Women 4 to 79 Years of Age (Panel A) and in the Use of Adjuvant Therapy among Women 5 to 69 Years of Age with Node-Positive Stage II or IIIA Breast Cancer (Panel B). The models used these data as common variables. In Panel B, the tamoxifen-only curve drops during the 199s because of the increased use of adjuvant chemotherapy for patients with estrogen-receptor positive tumors; the overall rate of use of tamoxifen, with and without chemotherapy, in the 199s is nearly constant at about 6 percent of patients, mostly those with estrogen-receptor positive tumors. The use of chemotherapy (calculated by adding the blue and red curves) increased from 55 percent to 87 percent during the same period. rate of death from breast cancer in the absence of screening or adjuvant treatment or both for a single model. We selected W for this demonstration because its results were typical. This figure shows that the estimated rate of death from breast cancer in the absence of screening and adjuvant therapy would have increased by about 3 percent from 1975 to 2. This increase derives from the modeled background trend in the incidence of breast cancer in the absence of screening. Table 3 gives the estimated reductions in the rate of death in 2 owing to the use of tamoxifen therapy, chemotherapy, both therapies combined, screening, and the overall combination. The decline in the rate of death from breast cancer in 2 attributable to the combination of screening and adjuvant therapy ranges from 24.9 percent to 38.3 percent, which because of the increasing background trend (that is, with no screening) is larger than the actual decline of 21.3 percent from 1975 to 2. However, in these models, the relative contributions of screening and treatment to the reduction in the rate of death from breast cancer are insensitive to variations in the background trend of the incidence of breast cancer. We estimated that screening as practiced in the United States reduced the rate of death from breast cancer in the range of 7 to 23 percent across the seven models, with a median of 15 percent. The percentage of the reduction attributable to adjuvant therapy was in the range of 12 to 21 percent, with a median of 19 percent. The combination of screening and adjuvant therapy reduced the rate of death by an estimated 25 to 38 percent, with a median of 3 percent. For each of the seven models, the combination of screening and adjuvant therapy reduced the rate of death by slightly less than the sum of the contributions from screening and adjuvant therapy alone. As indicated in Table 3, the proportion of the decrease in the rate of death from breast cancer that was attributable to screening ranged from 28 percent to 65 percent, with a median of 46 percent. Figure 3A shows a contour plot of the estimated distribution of a larger population of model results from which our seven models represent a sample. When considering the results of all seven models, the most likely conclusion is that the contributions of screening and adjuvant treatment are similar. The spread of the distribution shown in Figure 3B reflects the uncertainty that is present in the available data and the differences in the modeling approaches. 1789

The new england journal of medicine A Rate of Death from Breast Cancer (no./1, women) 6 5 4 3 2 1 G E S D M R W Actual U.S. rate 1975 198 1985 199 1995 2 Year B Rate of Death from Breast Cancer (no./1, women) 6 5 4 3 2 1 No screening or adjuvant therapy Screening only Adjuvant therapy only Screening and adjuvant therapy 1975 198 1985 199 1995 2 Year Figure 2. Estimated and Actual Rates of Death from Breast Cancer among Women 3 to 79 Years of Age from 1975 to 2 (Panel A) and under Hypothetical Assumptions about the Use of Screening Mammography and Adjuvant Treatment (Panel B). Panel A, which compares the model-based results with the actual rates in the United States from 1975 to 2, shows the variability across the model estimates. Some of the models were calibrated according to the observed rate of death from breast cancer in the United States, and some were not. Panel B shows the results from model W (the University of Wisconsin Madison) of estimated mortality trends for the four scenarios considered: no screening and no adjuvant treatment; base-case screening, but no adjuvant treatment; no screening, but base-case adjuvant treatment; base-case screening and adjuvant treatment. Rates in both panels are age-adjusted to the 2 U.S. standard. Table 3. Estimated Reductions in the Rate of Death from Breast Cancer in 2 Attributed to Adjuvant Treatments and Screening.* Tamoxifen Chemotherapy Both Therapies Screening Overall percent (percent of reduction) D (Dana Farber Cancer Institute) 6.1 6.1 12. (35) 22.7 (65) 32.9 E (Erasmus University Medical Center) 12. 9.6 2.9 (58) 15.3 (42) 3.9 G (Georgetown University) 7.7 7. 14.6 (54) 12.4 (46) 24.9 M (M.D. Anderson Cancer Center) 1.7 9.5 19.5 (65) 1.6 (35) 27.5 R (University of Rochester) NA NA 19. (72) 7.5 (28) 25.6 S (Stanford University) 8.9 6.9 14.9 (47) 16.9 (53) 29.9 W (University of Wisconsin Madison) 12.5 8.9 2.8 (51) 2.3 (49) 38.3 * Values are point estimates from each model; percentages in parentheses are the percentages of the overall reduction that are attributable to treatment or screening. NA denotes not applicable. discussion We present results from seven models that were developed to estimate the effect of screening mammography and adjuvant therapy on the rates of death from breast cancer from 1975 to 2 in the United States. The models used common sources of data, but their approaches and assumptions differed. Despite these differences, all seven groups concluded that screening and treatment have contributed to the observed decline in the rate of death from breast cancer and that the decline can be explained by a combination of screening and therapy and not by either one alone. 179

A 3 B Reduction in Rate of Death Attributed to Adjuvant Treatment (%) 25 2 15 1 5 R M G E S W D 3 25 Reduction in Rate of Death Attributed to Adjuvant Treatment (%) 2 15 1 5 5 1 15 2 25 3 Reduction in Rate of Death Attributed to Screening (%) 5 1 15 2 25 3 Reduction in Rate of Death Attributed to Screening (%) Figure 3. Estimated Joint Distribution of the Reduction in the Rate of Death from Breast Cancer among U.S. Women 3 to 79 Years of Age Attributed to Adjuvant Treatment and to Screening Mammography. Values are compared with the rate of death in 2 in the absence of both screening and treatment. Panel A shows the point estimates from the individual models (designated by their letters) provided in Table 3. The distribution contours for the combined model results are derived by kernel-density estimation; each contour shows the locus of points having a constant density. Each model s point estimate is assumed to be at the mean of its own bivariate normal density whose covariance structure was estimated from that of the seven model estimates. These seven densities were then averaged with equal weights to obtain an estimated posterior joint distribution. The hill in Panel B is a three-dimensional rendering of the contour plot. The height of the hill shows the likelihood of the corresponding reductions due to screening and treatment. For example, a point near the top of the hill (from 1 to 15 percent for screening and from 15 to 2 percent for treatment) is about twice as likely to be the actual state than is a point on the third largest contour in Panel A. Although the conclusions of these models are qualitatively similar, their estimates vary. This variability is not surprising, given the diversity of the modeling approaches and assumptions. Although we strove for some consistency in modeling by using common sources of data concerning the use of mammography and adjuvant therapy over time, rates of death from causes other than breast cancer, and trends in background incidence that were not due to screening, some groups involved in our consortium used equivalent input variables suiting their models structure. In addition, the groups used different data sets (surveillance information, results of meta-analyses of treatment trials, and data from screening trials) to estimate variables specific to their models. Some models focused on invasive breast cancer, and others accounted for ductal carcinoma in situ in the natural history of and screening for the disease. The differences in conclusions reflect uncertainties in the interpretation of available information, rather than contradictions among the models. The variability in the quantitative conclusions across the models (Fig. 3 and Table 3) demonstrates an interplay between screening and treatment. Screening would have no benefit if not followed by treatment (including surgery), and treatment is likely to be more effective if cancer is detected at earlier stages by screening. Because the increasing use of adjuvant therapies and screening occurred over nearly the same periods, distinguishing between the two effects is not easy. A consequence of the concurrent introduction of the two interventions is that slight variations in modeling assumptions can result in marked changes in estimated effects. It is reassuring that the qualitative conclusions agree in the face of this sensitivity to assumptions. The slight negative statistical correlation between the models estimates of the relative contributions of the two interventions (Fig. 3) reflects the intuitive notion that either screening or therapy can provide some of the benefit not offered by the other intervention. For example, if adjuvant therapy explains more of the observed reduction in mortality, then less is left for screening to explain. One of the models (model M) specifically addressed within- 1791

model uncertainty. The variability among models shown in Figure 3A is similar to the variability within the results of model M (not shown), suggesting that the variation among models is similar in magnitude to the variability within a single model s estimate after the uncertainties related to various aspects of the analysis are accounted for. Addressing a complex public health question through the collaboration of seven modeling groups is unusual in health and medical-decision sciences. The challenges of model comparisons were mitigated by this collaborative effort. Supported by grants (U1CA6374, U1CA8676, U1CA8682, U1CA63736, U1CA713, U1CA69976, U1CA63731, and U1CA74) from the National Cancer Institute to fund the Breast Cancer Surveillance Consortium and the Group Health Cooperative for information on breast-cancer screening practices and characteristics of tumors detected by screening and by cooperative agreements (CA88278, CA88211, CA8822, CA88248, CA88177, CA8827, and CA88283) with the National Cancer Institute. Dr. Berry reports having received consulting fees from Pfizer, Novartis, Eli Lilly, and Bristol-Myers Squibb. appendix The following were CISNET collaborators: M. Zelen (principal investigator), S.J. Lee, H. Huang, and R.S. Gelman, Dana Farber Cancer Institute, Boston; J.D.F. Habbema (principal investigator), S.Y.G.L. Tan, G.J. van Oortmarssen, H.J. de Koning, and R. Boer, Erasmus University Medical Center, Rotterdam, the Netherlands; J.S. Mandelblatt (principal investigator), W.F. Lawrence, B. Yi, J. Cullen, and K.R. Yabroff, Georgetown University, Washington, D.C.; C.B. Schechter, Albert Einstein College of Medicine, New York; D.A. Berry (principal investigator), L.Y. Inoue, M.F. Munsell, J. Venier, Y. Shen, G. Ball, E. Hoy, R.L. Theriault, and M.L. Bondy, M.D. Anderson Cancer Center, Houston; A.Y. Yakovlev (principal investigator), A.V. Zorin, and L.G. Hanin, University of Rochester, Rochester, N.Y.; S.K. Plevritis (principal investigator), B.M. Sigal, P. Salzman, P.W. Glynn, J. Rosenberg, and S. Rai, Stanford University, Stanford, Calif.; D.G. Fryback (principal investigator), M.A. Rosenberg, A. Trentham-Dietz, P.L. Remington, N.K. Stout, and V. Kuruchittham, University of Wisconsin Madison, Madison; E.J. Feuer (program director), K.A. Cronin, and A.B. Mariotto, National Cancer Institute, Bethesda, Md.; and L. Clarke, Cornerstone Systems, Lynden, Wash. references 1. Ries LAG, Eisner MP, Kosary CL, et al., eds. SEER cancer statistics review, 1975-21. Bethesda, Md.: National Cancer Institute, 24. 2. Parkin DM, Bray FI, Devesa SS. Corrigendum to Cancer burden in the year 2: the global picture [Eur J Cancer 21;37: Suppl 8:S4-S66]. Eur J Cancer 23;39: 848. 3. Tabar L, Duffy SW, Vitak B, Chen HH, Prevost TC. The natural history of breast carcinoma: what have we learned from screening? Cancer 1999;86:449-62. 4. Leitch AM, Dodd GD, Costanza M, et al. American Cancer Society guidelines for the early detection of breast cancer: update 1997. CA Cancer J Clin 1997;47:15-3. 5. Smith RA, Saslow D, Sawyer KA, et al. American Cancer Society guidelines for breast cancer screening: update 23. CA Cancer J Clin 23;53:141-69. 6. Humphrey LL, Helfand M, Chan BK, Woolf SH. Breast cancer screening: a summary of the evidence for the U.S. Preventive Services Task Force. Ann Intern Med 22; 137:347-6. 7. Swan J, Breen N, Coates RJ, Rimer BK, Lee NC. Progress in cancer screening practices in the United States: results from the 2 National Health Interview Survey. Cancer 23;97:1528-4. 8. Olsen O, Gøtzsche PC. Cochrane review on screening for breast cancer with mammography. Lancet 21;358:134-2. 9. Nyström L, Andersson I, Bjurstam N, Frisell J, Nordenskjöld B, Rutqvist LE. Longterm effects of mammography screening: updated overview of the Swedish randomised trials. Lancet 22;359:99-19. [Erratum, Lancet 22;36:724.] 1. Goodman SN. The mammography dilemma: a crisis for evidence-based medicine? Ann Intern Med 22;137:363-5. 11. Early Breast Cancer Trialists Collaborative Group. Polychemotherapy for early breast cancer: an overview of the randomised trials. Lancet 1998;352:93-42. 12. Idem. Multi-agent chemotherapy for early breast cancer. Cochrane Database Syst Rev 22;1:CD487. 13. Idem. Tamoxifen for early breast cancer: an overview of the randomised trials. Lancet 1998;351:1451-67. 14. Cancer Intervention and Surveillance ing Network home page. (Accessed September 3, 25, at http://cisnet.cancer.gov.) 15. Harlan LC, Abrams J, Warren JL, Clegg L, Stevens J, Ballard-Barbash R. Adjuvant therapy for breast cancer: practice patterns of community physicians. J Clin Oncol 22; 2:189-17. 16. Cancer Intervention and Surveillance ing Network: input parameter generator interfaces. (Accessed September 3, 25, at http://cisnet.cancer.gov/interfaces/.) 17. Breen N, Wagener DK, Brown ML, Davis WW, Ballard-Barbash R. Progress in cancer screening over a decade: results of cancer screening from the 1987, 1992, and 1998 National Health Interview Surveys. J Natl Cancer Inst 21;93:174-13. 18. Ballard-Barbash R, Taplin SH, Yankaskas BC, et al. Breast Cancer Surveillance Consortium: a national mammography screening and outcomes database. AJR Am J Roentgenol 1997;169:11-8. 19. Human Mortality Database. Berkeley: University of California, Berkeley. (Accessed September 3, 25, at http://www.mortality. org.) 2. Tabar L, Vitak B, Chen HH, et al. The Swedish Two-County Trial twenty years later: updated mortality results and new insights from long-term follow-up. Radiol Clin North Am 2;38:625-51. 21. Miller AB, To T, Baines CJ, Wall C. Canadian National Breast Screening Study-2: 13- year results of a randomized trial in women aged 5-59 years. J Natl Cancer Inst 2; 92:149-9. 22. Idem. Canadian National Breast Screening Study-1: breast cancer mortality after 11 to 16 years of follow-up: a randomized screening trial of mammography in women aged 4 to 49 years. Ann Intern Med 22; 137:35-12. 23. Cancer Intervention and Surveillance ing Network: model profiles. (Accessed September 3, 25, at http://cisnet. cancer.gov/resources/.) 24. Wand MP, Jones MC. Kernel smoothing. Boca Raton, Fla.: Chapman and Hall/ CRC Press, 1995. 25. Cronin KA, Yu B, Krapcho M, et al. ing the dissemination of mammography in the United States. Cancer Causes Control 25;16:71-12. 26. Mariotto A, Feuer EJ, Harlan LC, Wun LM, Johnson KA, Abrams J. Trends in use of adjuvant multi-agent chemotherapy and tamoxifen for breast cancer in the United States: 1975-1999. J Natl Cancer Inst 22; 94:1626-34. Copyright 25 Massachusetts Medical Society. 1792